Massive Mimo Scheduler Eval

massive-mimo-scheduler-eval

A Deep Reinforcement Learning-Based Resource Scheduler for Massive MIMO Networks — Qing An et al. (2023) (arXiv:2303.00958, 2023)

What this evaluates

Evaluates the ability of a deep reinforcement learning scheduler to allocate wireless resources (users to resource blocks) in massive MIMO networks. It probes the model's capacity to maximize spectral efficiency and user fairness under varying channel conditions (static vs. mobile) and network scales.

Datasets

• QuaDRiGa 3GPP_3D_UMi_LOS — total ?; splits: test (400)

Metrics

• normalized spectral efficiency (primary) — range: other
  • Average achievable data rate per resource block normalized by system bandwidth, computed over 400 test TTIs.
• JFI — range: [0, 1]
  • Jain's Fairness Index, measuring the fairness of resource allocation among users across the 400 test TTIs.

Input / output format

Input: Channel state information (CSI) represented either as a raw channel matrix or as user grouping labels derived from channel correlation.

Output: Discrete scheduling decisions indicating which users are assigned to each resource block (e.g., a scheduled user set U_b per resource block b).

Scoring recipe

# For each of 400 test TTIs:
# 1. Feed channel matrix/labels to scheduler
# 2. Get scheduled user set U_b for each resource block b
# 3. Compute spectral efficiency SE_t and JFI_t for TTI t
# 4. Average over all TTIs:
avg_se = sum(SE_t for t in 1..400) / 400
avg_jfi = sum(JFI_t for t in 1..400) / 400
return avg_se, avg_jfi

Common pitfalls

• Using outdated or partial channel information during testing, which severely degrades scheduler performance in mobile scenarios.
• Comparing schedulers across different network sizes without accounting for computational infeasibility of optimal baselines (Opt-PF/Opt-MR) at 64x64 scale.

Evidence (verbatim from paper)

In the testing phase, we run our simulation environment for additional 400 TTIs in the same cell and use the trained model to schedule users while recording the spectral efficiency and the JFI values across TTIs. For a fair comparison, we use the exact same channels generated as input to all benchmarks.

Citation

@misc{an2023deepreinforcementlearning,
  title={A Deep Reinforcement Learning-Based Resource Scheduler for Massive MIMO Networks},
  author={Qing An et al. (2023)},
  year={2023},
  note={arXiv:2303.00958}
}

• arXiv: 2303.00958